WiSER Model Checkpoint

WiSER is a wireless scene encoder for geometry-grounded radiomap and channel impulse response (CIR) prediction from sparse 3D indoor scenes.

This folder is the local Hugging Face model repository staging area for WiSER.

Project links:

• Project page: https://scp10086.github.io/wiser-page/
• Code: https://github.com/scp10086/WiSER
• Dataset: https://huggingface.co/datasets/Jingqiao-ucsc/sionna-scannetpp-small-100
• Paper: coming soon

Files

wiser_sparse_scene_encoder_small100_full.pt  final full checkpoint
config_snapshot.json                         training/model configuration snapshot
eval_summary.json                            validation metrics from the paper run

The checkpoint is released for the WiSER public code package. Some internal checkpoint keys may retain historical csi_* names for backward compatibility; the paper and public documentation use CIR.

Intended Use

Install the public WiSER code repository, then run:

python scripts/infer_example.py \
  --example-root example \
  --checkpoint /path/to/wiser_sparse_scene_encoder_small100_full.pt \
  --out-json outputs/example_summary.json

Full radiomap and CIR evaluation can be run with:

python scripts/evaluate_dual.py \
  --ckpt /path/to/wiser_sparse_scene_encoder_small100_full.pt \
  --d22-ckpt /path/to/wiser_sparse_scene_encoder_small100_full.pt \
  --radiomap-manifest /path/to/radiomap_manifest.json \
  --cir-manifest /path/to/cir_manifest.json \
  --wireless-root /path/to/wireless/scannetpp \
  --scene3d-root /path/to/processed/3D/scannetpp \
  --out-json outputs/eval_summary.json

Metrics

The bundled eval_summary.json records the paper checkpoint validation summary. Public documentation should report metrics from that file rather than from ad-hoc local probes.